Skip to content

README.md

Latest commit

 

History

History
17 lines (9 loc) · 1.17 KB

README.md

File metadata and controls

17 lines (9 loc) · 1.17 KB
Raw

vocal-imitation

A method for generating human-like vocal imitations of sounds.

Setting up the vocal synthesizer pipeline

The vocal tract model used to generate synthetic utterances is written in Faust, using faust-ctypes to import the compiled DLL into Python. First, get the latest Faust distribution (this project has been tested up to Faust 2.74.6).

The SF_voc_synth_m.dsp synth is the masculine speaker, and the SF_voc_synth_f.dsp synth is the feminine speaker.

Navigate to faust_dsp (cd faust_dsp), then compile the FAUST file to base C: faust -lang c -A faust_ctypes -a faust_ctypes/dllarch.c <DSP file> > <C file destination>.

Then, compile the to DLL using any C compiler: gcc -fPIC -shared <C file> -o SF_voc_synth_m.so.

Referent data

The FSD50k dataset is used as the set of referents. It's labeled using the AudioSet Ontology. Be sure to download all parts, then add the entire dataset to the data directory under data/FSD50/.